On Tue, Aug 01, 2017 at 01:58:40PM -0600, Jason Gunthorpe wrote: > On Tue, Aug 01, 2017 at 10:15:11AM -0500, Chien Tin Tung wrote: > > On Tue, Aug 01, 2017 at 09:22:47AM -0500, Christopher Lameter wrote: > > > On Tue, 1 Aug 2017, Chien Tin Tung wrote: > > > > > > > Why do I need to explain the original patch? It was accepted many kernels > > > > ago. Your questions on the original were based on false assumptions and > > > > facts which I've proven over and over. You are right in that I do > > > > not want to revisit those either. There is a patch that can solve > > > > the problem you are facing but yet you insist on the revert. This is > > > > very puzzling to me. You simply refuse to move this forward. That > > > > is your choice but revert is a no go for me. > > > > > > Ok this is pretty confusing to someone not involved in the prior > > > discussions. Could both of your stop attacking each other and start > > > talking about the technical issues? > > > > There's been multiple threads over this revert. This one is the > > latest Leon walked away from: https://patchwork.kernel.org/patch/9814367/ > > > > I'm simply asking for proof/evidence/facts to back up Leon's claims. > > Show me the code/stack trace/whatever and I will be happy to admit wrong > > publicly on the list and move on. I do have better things to do than > > to NACK a revert. > > Well, I think the design of the netlink protocol in iwpmd is pretty > strange, but audit netlink is the other major user of blocking > netlink_uncast to achive some reliability, so it should be workable > here as well. Agreed, somewhat. I'm all for making things better but the basis of any conversation needs to start with correct assumptions. > Presumably the iwpmd kernel side is designed to handle the locking > right to avoid deadlocks. Please stop using the word deadlock in relation to usage of 1-shot retry in ibnl_unicast. It does not and will not cause deadlock period. It will wait for a buffer within timeout value but that's it. > I'm not sure why there is so much noise about this - yes, iwpmd is > really weird, but it is a UAPI, we can't change it and we can't demand > they change. For better or worse the protocol is based on > 'near-reliable' netlink delivery of messages and we are stuck with > that. > > Leon is also right that every other user of netlink_unicast in rdma > should be using the non-blocking version. In Mustafa's patch, there are calls to both versions of ibnl_unicast. There are situations where the retry version of ibnl_unicast is not desirable, note I didn't say wrong. Here is the reality. The retry version of ibnl_unicast will only wait when there is no buffer to send the Netlink message. It does not interfere with other Netlink calls. For that matter, other calls to transmit a Netlink message will fail as there are no buffers available. If a buffer does not become available within timeout parameter, you have bigger problems. > So, the above patch seems like a sensible approach to me.. Finally. Thank you. Chien -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
